ray.rllib.policy.sample_batch.SampleBatch.__init__#

SampleBatch.__init__(*args, **kwargs)[source]#

Constructs a sample batch (same params as dict constructor).

Note: All args and those kwargs not listed below will be passed as-is to the parent dict constructor.

Parameters:
  • _time_major – Whether data in this sample batch is time-major. This is False by default and only relevant if the data contains sequences.

  • _max_seq_len – The max sequence chunk length if the data contains sequences.

  • _zero_padded – Whether the data in this batch contains sequences AND these sequences are right-zero-padded according to the _max_seq_len setting.

  • _is_training – Whether this batch is used for training. If False, batch may be used for e.g. action computations (inference).